SDMtune provides a user-friendly framework that enables the training and the evaluation of species distribution models (SDMs). The package implements functions for data driven variable selection and model tuning and includes numerous utilities to display the results. All the functions used to select variables or to tune model hyperparameters have an interactive real-time chart displayed in the RStudio viewer pane during their execution. Visit the package website and learn how to use SDMtune starting from the first article Prepare data for the analysis.
You can install the latest release version from CRAN:
Or the development version from GitHub:
SDMtune implements three functions for hyperparameters tuning:
gridSearch: runs all the possible combinations of predefined hyperparameters’ values;
randomSearch: randomly selects a fraction of the possible combinations of predefined hyperparameters’ values;
optimizeModel: uses a genetic algorithm that aims to optimize the given evaluation metric by combining the predefined hyperparameters’ values.
When the amount of hyperparameters’ combinations is high, the computation time necessary to train all the defined models could be very long. The function
optimizeModel offers a valid alternative that reduces computation time thanks to an implemented genetic algorithm. This function seeks the best combination of hyperparameters reaching a near optimal or optimal solution in a reduced amount of time compared to
gridSearch. The following code shows an example using a simulated dataset. First a model is trained using the Maxnet algorithm implemented in the
maxnet package with default hyperparameters’ values. After the model is trained, both the
optimizeModel functions are executed to compare the execution time and model performance evaluated with the AUC metric. If the following code is not clear, please check the articles in the website.
library(SDMtune) # Acquire environmental variables files <- list.files(path = file.path(system.file(package = "dismo"), "ex"), pattern = "grd", full.names = TRUE) predictors <- raster::stack(files) # Prepare presence and background locations p_coords <- virtualSp$presence bg_coords <- virtualSp$background # Create SWD object data <- prepareSWD(species = "Virtual species", p = p_coords, a = bg_coords, env = predictors, categorical = "biome") # Split presence locations in training (80%) and testing (20%) datasets datasets <- trainValTest(data, test = 0.2, only_presence = TRUE, seed = 25) train <- datasets[] test <- datasets[] # Train a Maxnet model model <- train(method = "Maxnet", data = train) # Define the hyperparameters to test h <- list(reg = seq(0.1, 3, 0.1), fc = c("lq", "lh", "lqp", "lqph", "lqpht")) # Test all the possible combinations with gridSearch gs <- gridSearch(model, hypers = h, metric = "auc", test = test) head(gs@results[order(-gs@results$test_AUC), ]) # Best combinations # Use the genetic algorithm instead with optimizeModel om <- optimizeModel(model, hypers = h, metric = "auc", test = test, seed = 4) head(om@results) # Best combinations
During the execution of “tuning” and “variable selection” functions, real-time charts displaying training and validation metrics are displayed in the RStudio viewer pane (below is a screencast of the previous executed
In the following example we train a Maxent model:
# Train a Maxent model sdmtune_model <- train(method = "Maxent", data = data)
We compare the execution time of the
predict function between SDMtune that uses its own algorithm and dismo (Hijmans et al. 2017) that calls the MaxEnt Java software (Phillips, Anderson, and Schapire 2006). We first convert the object
sdmtune_model in a object that is accepted by dismo:
maxent_model <- SDMmodel2MaxEnt(sdmtune_model)
Next is a function used below to test if the results are equal, with a tolerance of
Now we test the execution time using the microbenckmark package:
and plot the output:
library(ggplot2) ggplot(bench, aes(x = expr, y = time/1000000, fill = expr)) + geom_boxplot() + labs(fill = "", x = "Package", y = "time (milliseconds)") + theme_minimal()
To train a Maxent model using the Java implementation you need that:
The file maxent.jar can be downloaded here (note that you need MaxEnt version >= 3.4.1 (Phillips et al. 2017)). This file must be copied into the right folder to be available for the
dismo package (Hijmans et al. 2017): copy the file maxent.jar into the folder named java that is located inside the folder returned by the following command:
checkMaxentInstallation checks that Java JDK and rJava are installed, and that the file maxent.jar is in the correct folder.
If everything is correctly configured for
dismo, the following command will return the used MaxEnt version (make sure that the version is >= 3.4.1):
Please note that this project follows a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.
Hijmans, Robert J., Steven Phillips, John Leathwick, and Jane Elith.
Phillips, Steven J., Robert P. Anderson, Miroslav Dudík, Robert E. Schapire, and Mary E. Blair. 2017. “Opening the black box: an open-source release of Maxent.” Ecography 40 (7): 887–93. https://doi.org/10.1111/ecog.03049.
Phillips, Steven J, Robert P Anderson, and Robert E Schapire. 2006. “Maximum entropy modeling of species geographic distributions.” Ecological Modelling 190: 231–59. https://doi.org/10.1016/j.ecolmodel.2005.03.026.